Merge branch 'comfyanonymous:master' into deon-mac

.github/workflows/windows_release_nightly_pytorch.yml

@@ -7,19 +7,19 @@ on:
         description: 'cuda version'
         required: true
         type: string
-        default: "124"
+        default: "126"
 
       python_minor:
         description: 'python minor version'
         required: true
         type: string
-        default: "12"
+        default: "13"
 
       python_patch:
         description: 'python patch version'
         required: true
         type: string
-        default: "4"
+        default: "1"
 #  push:
 #    branches:
 #      - master

api_server/services/terminal_service.py

@@ -28,7 +28,7 @@ def update_size(self):
 
         if columns != self.cols:
             self.cols = columns
-            changed = True 
+            changed = True
 
         if lines != self.rows:
             self.rows = lines

app/custom_node_manager.py

@@ -10,7 +10,7 @@ class CustomNodeManager:
     Placeholder to refactor the custom node management features from ComfyUI-Manager.
     Currently it only contains the custom workflow templates feature.
     """
-    def add_routes(self, routes, webapp, loadedModules):     
+    def add_routes(self, routes, webapp, loadedModules):
 
         @routes.get("/workflow_templates")
         async def get_workflow_templates(request):

comfy/comfy_types/README.md

@@ -5,7 +5,7 @@ This module provides type hinting and concrete convenience types for node develo
 If cloned to the custom_nodes directory of ComfyUI, types can be imported using:
 
 ```python
-from comfy_types import IO, ComfyNodeABC, CheckLazyMixin
+from comfy.comfy_types import IO, ComfyNodeABC, CheckLazyMixin
 
 class ExampleNode(ComfyNodeABC):
     @classmethod

comfy/comfy_types/examples/example_nodes.py

@@ -1,12 +1,12 @@
-from comfy_types import IO, ComfyNodeABC, InputTypeDict
+from comfy.comfy_types import IO, ComfyNodeABC, InputTypeDict
 from inspect import cleandoc
 
 
 class ExampleNode(ComfyNodeABC):
     """An example node that just adds 1 to an input integer.
 
-    * Requires an IDE configured with analysis paths etc to be worth looking at.
-    * Not intended for use in ComfyUI.
+    * Requires a modern IDE to provide any benefit (detail: an IDE configured with analysis paths etc).
+    * This node is intended as an example for developers only.
     """
 
     DESCRIPTION = cleandoc(__doc__)

comfy/extra_samplers/uni_pc.py

@@ -226,7 +226,7 @@ def model_wrapper(
             The input `model` has the following format:
             ``
                 model(x, t_input, **model_kwargs) -> noise | x_start | v | score
-            `` 
+            ``
 
             The input `classifier_fn` has the following format:
             ``
@@ -240,7 +240,7 @@ def model_wrapper(
             The input `model` has the following format:
             ``
                 model(x, t_input, cond, **model_kwargs) -> noise | x_start | v | score
-            `` 
+            ``
             And if cond == `unconditional_condition`, the model output is the unconditional DPM output.
 
             [4] Ho, Jonathan, and Tim Salimans. "Classifier-free diffusion guidance."
@@ -254,7 +254,7 @@ def model_wrapper(
     ``
         def model_fn(x, t_continuous) -> noise:
             t_input = get_model_input_time(t_continuous)
-            return noise_pred(model, x, t_input, **model_kwargs)         
+            return noise_pred(model, x, t_input, **model_kwargs)
     ``
     where `t_continuous` is the continuous time labels (i.e. epsilon to T). And we use `model_fn` for DPM-Solver.
 
@@ -359,7 +359,7 @@ def __init__(
         max_val=1.,
         variant='bh1',
     ):
-        """Construct a UniPC. 
+        """Construct a UniPC.
 
         We support both data_prediction and noise_prediction.
         """
@@ -372,7 +372,7 @@ def __init__(
 
     def dynamic_thresholding_fn(self, x0, t=None):
         """
-        The dynamic thresholding method. 
+        The dynamic thresholding method.
         """
         dims = x0.dim()
         p = self.dynamic_thresholding_ratio
@@ -404,7 +404,7 @@ def data_prediction_fn(self, x, t):
 
     def model_fn(self, x, t):
         """
-        Convert the model to the noise prediction model or the data prediction model. 
+        Convert the model to the noise prediction model or the data prediction model.
         """
         if self.predict_x0:
             return self.data_prediction_fn(x, t)
@@ -461,7 +461,7 @@ def get_orders_and_timesteps_for_singlestep_solver(self, steps, order, skip_type
 
     def denoise_to_zero_fn(self, x, s):
         """
-        Denoise at the final step, which is equivalent to solve the ODE from lambda_s to infty by first-order discretization. 
+        Denoise at the final step, which is equivalent to solve the ODE from lambda_s to infty by first-order discretization.
         """
         return self.data_prediction_fn(x, s)
 
@@ -510,7 +510,7 @@ def multistep_uni_pc_vary_update(self, x, model_prev_list, t_prev_list, t, order
         col = torch.ones_like(rks)
         for k in range(1, K + 1):
             C.append(col)
-            col = col * rks / (k + 1) 
+            col = col * rks / (k + 1)
         C = torch.stack(C, dim=1)
 
         if len(D1s) > 0:
@@ -626,7 +626,7 @@ def multistep_uni_pc_bh_update(self, x, model_prev_list, t_prev_list, t, order,
             R.append(torch.pow(rks, i - 1))
             b.append(h_phi_k * factorial_i / B_h)
             factorial_i *= (i + 1)
-            h_phi_k = h_phi_k / hh - 1 / factorial_i 
+            h_phi_k = h_phi_k / hh - 1 / factorial_i
 
         R = torch.stack(R)
         b = torch.tensor(b, device=x.device)

comfy/ldm/cascade/stage_b.py

@@ -138,7 +138,7 @@ def get_block(block_type, c_hidden, nhead, c_skip=0, dropout=0, self_attn=True):
     #     nn.init.normal_(self.pixels_mapper[2].weight, std=0.02)  # conditionings
     #     torch.nn.init.xavier_uniform_(self.embedding[1].weight, 0.02)  # inputs
     #     nn.init.constant_(self.clf[1].weight, 0)  # outputs
-    # 
+    #
     #     # blocks
     #     for level_block in self.down_blocks + self.up_blocks:
     #         for block in level_block:
@@ -148,7 +148,7 @@ def get_block(block_type, c_hidden, nhead, c_skip=0, dropout=0, self_attn=True):
     #                 for layer in block.modules():
     #                     if isinstance(layer, nn.Linear):
     #                         nn.init.constant_(layer.weight, 0)
-    # 
+    #
     # def _init_weights(self, m):
     #     if isinstance(m, (nn.Conv2d, nn.Linear)):
     #         torch.nn.init.xavier_uniform_(m.weight)

comfy/ldm/cascade/stage_c.py

@@ -142,7 +142,7 @@ def get_block(block_type, c_hidden, nhead, c_skip=0, dropout=0, self_attn=True):
     #     nn.init.normal_(self.clip_img_mapper.weight, std=0.02)  # conditionings
     #     torch.nn.init.xavier_uniform_(self.embedding[1].weight, 0.02)  # inputs
     #     nn.init.constant_(self.clf[1].weight, 0)  # outputs
-    # 
+    #
     #     # blocks
     #     for level_block in self.down_blocks + self.up_blocks:
     #         for block in level_block:
@@ -152,7 +152,7 @@ def get_block(block_type, c_hidden, nhead, c_skip=0, dropout=0, self_attn=True):
     #                 for layer in block.modules():
     #                     if isinstance(layer, nn.Linear):
     #                         nn.init.constant_(layer.weight, 0)
-    # 
+    #
     # def _init_weights(self, m):
     #     if isinstance(m, (nn.Conv2d, nn.Linear)):
     #         torch.nn.init.xavier_uniform_(m.weight)

comfy/ldm/flux/model.py

@@ -168,7 +168,7 @@ def block_wrap(args):
                 out = blocks_replace[("single_block", i)]({"img": img,
                                                            "vec": vec,
                                                            "pe": pe,
-                                                           "attn_mask": attn_mask}, 
+                                                           "attn_mask": attn_mask},
                                                           {"original_block": block_wrap})
                 img = out["img"]
             else:

comfy/ldm/hydit/attn_layers.py

@@ -159,7 +159,7 @@ def forward(self, x, y, freqs_cis_img=None):
 
         q = q.transpose(-2, -3).contiguous()        # q ->  B, L1, H, C - B, H, L1, C
         k = k.transpose(-2, -3).contiguous()      # k ->  B, L2, H, C - B, H, C, L2
-        v = v.transpose(-2, -3).contiguous() 
+        v = v.transpose(-2, -3).contiguous()
 
         context = optimized_attention(q, k, v, self.num_heads, skip_reshape=True, attn_precision=self.attn_precision)
 

comfy/ldm/modules/sub_quadratic_attention.py

@@ -17,10 +17,10 @@
 import logging
 
 try:
-	from typing import Optional, NamedTuple, List, Protocol
+    from typing import Optional, NamedTuple, List, Protocol
 except ImportError:
-	from typing import Optional, NamedTuple, List
-	from typing_extensions import Protocol
+    from typing import Optional, NamedTuple, List
+    from typing_extensions import Protocol
 
 from typing import List
 

comfy/model_base.py

@@ -787,7 +787,7 @@ def extra_conds(self, **kwargs):
         cross_attn = kwargs.get("cross_attn", None)
         if cross_attn is not None:
             out['c_crossattn'] = comfy.conds.CONDRegular(cross_attn)
-        # upscale the attention mask, since now we 
+        # upscale the attention mask, since now we
         attention_mask = kwargs.get("attention_mask", None)
         if attention_mask is not None:
             shape = kwargs["noise"].shape

comfy/model_detection.py

@@ -576,7 +576,7 @@ def unet_config_from_diffusers_unet(state_dict, dtype=None):
             'dtype': dtype, 'in_channels': 9, 'model_channels': 320, 'num_res_blocks': [2, 2, 2, 2], 'transformer_depth': [1, 1, 1, 1, 1, 1, 0, 0],
             'channel_mult': [1, 2, 4, 4], 'transformer_depth_middle': 1, 'use_linear_in_transformer': False, 'context_dim': 768, 'num_heads': 8,
             'transformer_depth_output': [1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0],
-            'use_temporal_attention': False, 'use_temporal_resblock': False}  
+            'use_temporal_attention': False, 'use_temporal_resblock': False}
 
 
     supported_models = [SDXL, SDXL_refiner, SD21, SD15, SD21_uncliph, SD21_unclipl, SDXL_mid_cnet, SDXL_small_cnet, SDXL_diffusers_inpaint, SSD_1B, Segmind_Vega, KOALA_700M, KOALA_1B, SD09_XS, SD_XS, SDXL_diffusers_ip2p, SD15_diffusers_inpaint]

comfy/model_management.py

@@ -1121,9 +1121,8 @@ def soft_empty_cache(force=False):
     elif is_ascend_npu():
         torch.npu.empty_cache()
     elif torch.cuda.is_available():
-        if force or is_nvidia(): #This seems to make things worse on ROCm so I only do it for cuda
-            torch.cuda.empty_cache()
-            torch.cuda.ipc_collect()
+        torch.cuda.empty_cache()
+        torch.cuda.ipc_collect()
 
 def unload_all_models():
     free_memory(1e30, get_torch_device())

comfy/samplers.py

@@ -1,12 +1,13 @@
 from __future__ import annotations
 from .k_diffusion import sampling as k_diffusion_sampling
 from .extra_samplers import uni_pc
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Callable, NamedTuple
 if TYPE_CHECKING:
     from comfy.model_patcher import ModelPatcher
     from comfy.model_base import BaseModel
     from comfy.controlnet import ControlBase
 import torch
+from functools import partial
 import collections
 from comfy import model_management
 import math
@@ -224,7 +225,7 @@ def _calc_cond_batch(model: 'BaseModel', conds: list[list[dict]], x_in: torch.Te
         default_conds.append(default_c)
 
     if has_default_conds:
-        finalize_default_conds(model, hooked_to_run, default_conds, x_in, timestep)
+        finalize_default_conds(model, hooked_to_run, default_conds, x_in, timestep, model_options)
 
     model.current_patcher.prepare_state(timestep)
 
@@ -920,31 +921,37 @@ def sample(model, noise, positive, negative, cfg, device, sampler, sigmas, model
     return cfg_guider.sample(noise, latent_image, sampler, sigmas, denoise_mask, callback, disable_pbar, seed)
 
 
-SCHEDULER_NAMES = ["normal", "karras", "exponential", "sgm_uniform", "simple", "ddim_uniform", "beta", "linear_quadratic", "kl_optimal"]
 SAMPLER_NAMES = KSAMPLER_NAMES + ["ddim", "uni_pc", "uni_pc_bh2"]
 
-def calculate_sigmas(model_sampling, scheduler_name, steps):
-    if scheduler_name == "karras":
-        sigmas = k_diffusion_sampling.get_sigmas_karras(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
-    elif scheduler_name == "exponential":
-        sigmas = k_diffusion_sampling.get_sigmas_exponential(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
-    elif scheduler_name == "normal":
-        sigmas = normal_scheduler(model_sampling, steps)
-    elif scheduler_name == "simple":
-        sigmas = simple_scheduler(model_sampling, steps)
-    elif scheduler_name == "ddim_uniform":
-        sigmas = ddim_scheduler(model_sampling, steps)
-    elif scheduler_name == "sgm_uniform":
-        sigmas = normal_scheduler(model_sampling, steps, sgm=True)
-    elif scheduler_name == "beta":
-        sigmas = beta_scheduler(model_sampling, steps)
-    elif scheduler_name == "linear_quadratic":
-        sigmas = linear_quadratic_schedule(model_sampling, steps)
-    elif scheduler_name == "kl_optimal":
-        sigmas = kl_optimal_scheduler(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
-    else:
-        logging.error("error invalid scheduler {}".format(scheduler_name))
-    return sigmas
+class SchedulerHandler(NamedTuple):
+    handler: Callable[..., torch.Tensor]
+    # Boolean indicates whether to call the handler like:
+    #  scheduler_function(model_sampling, steps) or
+    #  scheduler_function(n, sigma_min: float, sigma_max: float)
+    use_ms: bool = True
+
+SCHEDULER_HANDLERS = {
+    "normal": SchedulerHandler(normal_scheduler),
+    "karras": SchedulerHandler(k_diffusion_sampling.get_sigmas_karras, use_ms=False),
+    "exponential": SchedulerHandler(k_diffusion_sampling.get_sigmas_exponential, use_ms=False),
+    "sgm_uniform": SchedulerHandler(partial(normal_scheduler, sgm=True)),
+    "simple": SchedulerHandler(simple_scheduler),
+    "ddim_uniform": SchedulerHandler(ddim_scheduler),
+    "beta": SchedulerHandler(beta_scheduler),
+    "linear_quadratic": SchedulerHandler(linear_quadratic_schedule),
+    "kl_optimal": SchedulerHandler(kl_optimal_scheduler, use_ms=False),
+}
+SCHEDULER_NAMES = list(SCHEDULER_HANDLERS)
+
+def calculate_sigmas(model_sampling: object, scheduler_name: str, steps: int) -> torch.Tensor:
+    handler = SCHEDULER_HANDLERS.get(scheduler_name)
+    if handler is None:
+        err = f"error invalid scheduler {scheduler_name}"
+        logging.error(err)
+        raise ValueError(err)
+    if handler.use_ms:
+        return handler.handler(model_sampling, steps)
+    return handler.handler(n=steps, sigma_min=float(model_sampling.sigma_min), sigma_max=float(model_sampling.sigma_max))
 
 def sampler_object(name):
     if name == "uni_pc":

comfy/utils.py

@@ -727,7 +727,7 @@ def slerp(b1, b2, r):
         res *= (b1_norms * (1.0-r) + b2_norms * r).expand(-1,c)
 
         #edge cases for same or polar opposites
-        res[dot > 1 - 1e-5] = b1[dot > 1 - 1e-5] 
+        res[dot > 1 - 1e-5] = b1[dot > 1 - 1e-5]
         res[dot < 1e-5 - 1] = (b1 * (1.0-r) + b2 * r)[dot < 1e-5 - 1]
         return res
 
@@ -893,7 +893,7 @@ def mult_list_upscale(a):
         out = torch.zeros([s.shape[0], out_channels] + mult_list_upscale(s.shape[2:]), device=output_device)
         out_div = torch.zeros([s.shape[0], out_channels] + mult_list_upscale(s.shape[2:]), device=output_device)
 
-        positions = [range(0, s.shape[d+2], tile[d] - overlap[d]) if s.shape[d+2] > tile[d] else [0] for d in range(dims)]
+        positions = [range(0, s.shape[d+2] - overlap[d], tile[d] - overlap[d]) if s.shape[d+2] > tile[d] else [0] for d in range(dims)]
 
         for it in itertools.product(*positions):
             s_in = s

comfy_extras/nodes_gits.py

@@ -162,7 +162,7 @@ def loglinear_interp(t_steps, num_steps):
         [14.61464119, 7.49001646, 5.85520077, 4.45427561, 3.46139455, 2.84484982, 2.19988537, 1.72759056, 1.36964464, 1.08895338, 0.86115354, 0.69515091, 0.54755926, 0.43325692, 0.34370604, 0.25053367, 0.17026083, 0.09824532, 0.02916753],
         [14.61464119, 11.54541874, 7.49001646, 5.85520077, 4.45427561, 3.46139455, 2.84484982, 2.19988537, 1.72759056, 1.36964464, 1.08895338, 0.86115354, 0.69515091, 0.54755926, 0.43325692, 0.34370604, 0.25053367, 0.17026083, 0.09824532, 0.02916753],
         [14.61464119, 11.54541874, 7.49001646, 5.85520077, 4.45427561, 3.46139455, 2.84484982, 2.19988537, 1.72759056, 1.36964464, 1.08895338, 0.89115214, 0.72133851, 0.59516323, 0.4783645, 0.38853383, 0.29807833, 0.22545385, 0.17026083, 0.09824532, 0.02916753],
-    ], 
+    ],
     1.15: [
         [14.61464119, 0.83188516, 0.02916753],
         [14.61464119, 1.84880662, 0.59516323, 0.02916753],
@@ -246,7 +246,7 @@ def loglinear_interp(t_steps, num_steps):
         [14.61464119, 5.85520077, 2.84484982, 1.72759056, 1.162866, 0.83188516, 0.64427125, 0.52423614, 0.43325692, 0.36617002, 0.32104823, 0.27464288, 0.25053367, 0.22545385, 0.19894916, 0.17026083, 0.13792117, 0.09824532, 0.02916753],
         [14.61464119, 5.85520077, 2.84484982, 1.78698075, 1.24153244, 0.92192322, 0.72133851, 0.57119018, 0.45573691, 0.38853383, 0.34370604, 0.29807833, 0.27464288, 0.25053367, 0.22545385, 0.19894916, 0.17026083, 0.13792117, 0.09824532, 0.02916753],
         [14.61464119, 5.85520077, 2.84484982, 1.78698075, 1.24153244, 0.92192322, 0.72133851, 0.57119018, 0.4783645, 0.41087446, 0.36617002, 0.32104823, 0.29807833, 0.27464288, 0.25053367, 0.22545385, 0.19894916, 0.17026083, 0.13792117, 0.09824532, 0.02916753],
-    ], 
+    ],
     1.35: [
         [14.61464119, 0.69515091, 0.02916753],
         [14.61464119, 0.95350921, 0.34370604, 0.02916753],